One thing that differentiates policy-driven machine learning is that new public policies are often implemented in a trial-and-error fashion, as data might not be available upfront. In this work, we try to accomplish approximate group fairness in an online decision-making process where examples are sampled \textit{i.i.d} from an underlying distribution. Our work follows from the classical learning-from-experts scheme, extending the multiplicative weights algorithm by keeping separate weights for label classes as well as groups. Although accuracy and fairness are often conflicting goals, we try to mitigate the trade-offs using an optimization step and demonstrate the performance on real data set.